Daniel E. Dombroski and John P. Crimaldi. The accuracy of acoustic

Transcription

Daniel E. Dombroski and John P. Crimaldi. The accuracy of acoustic
LIMNOLOGY
and
OCEANOGRAPHY: METHODS
Limnol. Oceanogr.: Methods 5, 2007, 23–33
© 2007, by the American Society of Limnology and Oceanography, Inc.
The accuracy of acoustic Doppler velocimetry measurements in
turbulent boundary layer flows over a smooth bed
Daniel E. Dombroski and John P. Crimaldi*
Department of Civil and Environmental Engineering, University of Colorado at Boulder, 428 UCB, Boulder, CO 80309-0428
Abstract
An evaluation of acoustic Doppler velocimetry (ADV) in the near-bed region of turbulent boundary layer
flows is presented. Acoustic instruments have large sampling volumes compared with the smallest scales of
motion in turbulent flows. This limits the accuracy of the technique, especially when making measurements
close to the bed or in flows where large spatial gradients are present. These limitations are quantified by comparing ADV results to laser Doppler velocimetry (LDV) measurements from the same flow, and to direct numerical simulations of similar flows. A SonTek 16 MHz ADVField system was used in the evaluation. Measurements
were made in a turbulent boundary layer over a smooth bed in a laboratory flume. The instrument was evaluated in both a fast and slow flow case with free stream velocities similar to those found in many natural environments. Additionally, results from an assessment of ADV sample volume size and position are reported. Mean
velocities were within 5% accuracy down to 0.7 cm from the bed for the fast flow case and 1 cm for the slow
flow case. ADV-derived Reynolds stresses matched well with those from LDV to within 1-2 cm of the bed, however turbulence intensities were found to deviate markedly up to 3-4 cm from the bed in some cases. In general,
errors are largest close to the bed, but extend further away from the bed than reported in previous studies.
Acoustic Doppler velocimetry (ADV) is a popular technique
for quantifying turbulent fluid flows in aquatic, marine, and
laboratory environments. The technique relies on the Doppler
shift principle to measure the velocity of suspended scattering
particles that are assumed to move passively with the flow.
Descriptions of ADV principles of operation are given by
Lohrmann et al. (1994), Sontek (1997), and Voulgaris and
Trowbridge (1998). The velocimeter employed in this evaluation is a SonTek 16 MHz MicroADV with the ADVField signal
processing hardware enclosed in a splash-proof housing. This
is a newer generation instrument that is advertised as having
better spatial and temporal characteristics than the original
SonTek 10 MHz and 5 MHz Ocean probes. These attributes
make the 16 MHz model ideal for laboratory and turbulence
measurements; however all of the models are available with
ADVField processors enabling the probe to be used in the field.
Acoustic instruments have a sampling volume (located at the
*
Corresponding author. E-mail: [email protected]
Acknowledgments
This research was supported by the National Science Foundation
(Grant IOB-0131553). The project benefited greatly from background
research conducted by Erin Carlson. We are also grateful for the
thoughtful comments of three anonymous reviewers.
intersection of the transmitting and receiving beams) that is
large compared to the smallest scales of motion in turbulent
flows. This limits the resolution and signal-to-noise (SNR) performance of the instrument, particularly in flows with smallscale turbulence and large velocity gradients. Thus, the ADV
technique is subject to certain limitations in different flow
regions and conditions. In this study, we evaluate ADV system
performance by using the instrument to measure controlled
laboratory flows. ADV results are compared to laser Doppler
velocimetry (LDV) measurements in the same flow and to
direct numerical simulations (DNS) of similar flows. The mean
flow rates are designed to mimic those typically found in natural marine and aquatic systems so as to provide a useful guide
for users interested in field and laboratory study.
Acoustic Doppler velocimetry—ADV is often used in studies of
turbulent processes in limnology (e.g., Cornelisen and
Thomas 2004; Tritico and Hotchkiss 2005; Venditti and Bauer
2005), oceanography (e.g., Yahel et al. 2002; Talke and Stacey
2003; Butt et al. 2004), and benthic ecology (e.g., Eckman et al.
1990; Friedrichs et al. 2000). The technique has been
employed in a range of flow conditions, including natural
stream environments (Lane et al. 1998; Mutz 2000; Carollo et
al. 2002), estuaries (Kim et al. 2000), natural and artificial
wave environments (Osborne et al. 1997; Doering and Baryla
2002), and artificial channels (Cheng and Chiew 1998; Car-
23
Dombroski and Crimaldi
Accuracy of ADV measurements
bonneau and Bergeron 2000; Nikora and Goring 2000; Lawless
and Robert 2001). ADV is typically used to measure mean
velocities, the strength of the velocity fluctuations away from
the mean (turbulence intensities), and temporal correlations
between the directional components of these fluctuations
(turbulent Reynolds stresses).
Laser Doppler velocimetry—The LDV technique (see Adrian
and Thompson 1993; Albrecht et al. 2003) is similar to ADV in
that it relies on the Doppler shift principle to measure the
velocity of suspended scattering particles in the flow. However, laser Doppler instruments have smaller sampling volumes than acoustic Doppler instruments due to the fact that
light waves are much shorter in wavelength than sound
waves. LDV is a commonly used technique for the characterization of boundary layer turbulence, primarily because the
measurement volume is nonintrusive and small relative to the
scales of motion in turbulent flows. This allows the user to
make highly spatially and temporally resolved measurements
throughout the boundary layer, including the near-bed region.
While generally not practical for field use, LDV is used extensively to examine turbulent flow in laboratory flumes (e.g.,
Nezu and Rodi 1986; Crimaldi et al. 2002).
This study adds to a collection of work from researchers
reporting on the accuracy and performance characteristics of
ADV systems. Early evaluations of ADV directly compared the
technique with results from LDV (Kraus et al 1994; Lohrmann
et al 1994), although this was a mostly qualitative assessment
of its performance. While these comparisons generally showed
favorable agreement, the techniques were not tested in challenging conditions, such as in the near-bed region, and there
was no objective comparison with known flow statistics. A
similar study conducted by Voulgaris and Trowbridge (1998)
produced a “ground-truth” estimate of true flow characteristics in a water flume. The estimate was derived from both ADV
(10 MHz model) and LDV measurements and was used as a
baseline of comparison for the instruments. The ADV measurements compared favorably with the “ground-truth” estimate, however the LDV measurements differed significantly.
Most of the data were above 10 mm from the bed, although
one experiment set that did include data points as close as 3 mm
to the bed suffered from a failure of the LDV instrument and,
therefore, direct comparison and ground-truthing was
unavailable. Finelli et al. (1999) extended the evaluation of
ADV to the near-bed region by comparing the technique to hotfilm velocimetry. It was shown that accurate knowledge of the
sample volume size and location is necessary for making measurements near the bed. Good agreement between the two
instruments (within 5%) was found above 10 mm from the
bed, however, below this point ADV velocities were as much
as 80% lower. Similar to other reports on accuracy of ADV, the
comparison gave an evaluation relative to another measurement technique without an objective method of independent evaluation. Furthermore, turbulent fluctuations were
not included in the hot-film velocimetry results. The authors
24
expressed the need for further research to be done in evaluating performance within 1 cm of the bed.
In this paper, we present direct comparisons of ADV and
LDV data in the same flow. We also provide DNS results of
comparable flows by Spalart (1988). While the DNS cannot
capture all the finer details of our laboratory flow, the agreement between the theoretical predictions and our LDV measurements is shown to generally be excellent. This provides
additional confidence for the LDV results to be used as an
independent baseline in evaluating ADV performance. The
aim of the study is to provide an evaluation of ADV at flow
rates typical of what might be found in natural environments,
including an evaluation of data from the near-bed region. An
independent evaluation of measurement volume size and
location was also performed, which allowed for precise alignment between the ADV sampling volume, the bed of the flume,
and the LDV. The results of this paper will help researchers to
understand the accuracy of acoustic Doppler systems, especially within a turbulent boundary layer.
Error sources—Before proceeding to our study, we first
review error sources related to the ADV technique. ADV is subject to several sources of error related to the size of the measurement volume and the nature of the acoustic system.
Firstly, positioning errors of the sampling volume can result in
interpretation errors in both turbulence statistics and mean
velocities when measuring flows in the near-bed region.
Finelli et al. (1999) developed a method for mapping the vertical size and position of the sample volume by systematically
moving the instrument vertically over an acoustic target and
quantifying the signal response. The results of the mapping
procedure found the sample volume to be significantly larger
(12.5-15.8 mm, depending on setup) than reported by the
manufacturer, and their determination of the sample volume
position differed from that reported by about 4 mm. These discrepancies are especially problematic when making measurements near the bed where large gradients are present.
Another source of error in ADV measurements is Doppler
noise, which is intrinsic to the technique and can add a
significant positive bias to the velocity power spectrum
(Lohrmann et al. 1994). This “white noise” can be identified
as an added noise floor in the high frequency region of the
spectrum. Typically, Doppler noise does not bias the mean
velocity measurements or the Reynolds stresses (as long as the
noise is uncorrelated between the channels upon which the
statistic is based and the signal strength is equal between the
receiving beams). However, the effects do change the unidirectional statistics associated with the fluctuating time series
(e.g., turbulence intensities). Increased spatial averaging can
reduce this type of noise, however the corresponding loss in
resolution may offset any improvements in accuracy. Thus,
the instrument is typically set up with a nominal sample volume size that represents an optimization between an acceptable noise level and the ability to resolve pertinent spatial
scales. Voulgaris and Trowbridge (1998) compared two meth-
Dombroski and Crimaldi
ods of removing noise from ADV data in post-processing. The
first method was spectral analysis, removing the noise tail in
the power spectrum. The second method used the results of
the ground-truthing technique but was potentially biased
high due to inclusion of errors related to the ADV and LDV
techniques. Voulgaris and Trowbridge (1998) also identify
errors resulting from limitations in the ability of the system to
resolve the Doppler shift of the acoustic pulse and errors due
to mean velocity shear within the sample volume.
Materials and procedures
Laboratory facility—The experiments were performed in a
recirculating water flume at the University of Colorado. The
flume test section features a glass bed and walls and is 15 m in
length and 1.25 m across. A stainless steel rod that trips the
turbulent boundary layer is mounted 0.30 m downstream
from the beginning of the test section. At the end of the test
section, different size weir plates are used to adjust the running depth of the flow. Two digitally controlled pumps ensure
accurate and repeatable flow conditions.
Acoustic Doppler velocimeter—The SonTek 16 MHz MicroADV
system used in this study has a 40-cm down-looking threedimensional (3D) stem. The manufacturer specifies a nominal
sample volume height of 9 mm and a distance of 50 mm from
the transmitter to the center of the sample volume. The system components include the probe, signal processing hardware, and communication and power cables. The processing
module is available as either the ADVLab model (PC card) or
the ADVField model (splash-proof housing or submersible
canister). Any probe of a given frequency can be used with a
processing module of that frequency, provided that the cables
and connectors are compatible (SonTek 1997). The system was
operated at a 10 Hz sampling rate using firmware version 8.04.
The ADV cannot resolve frequencies above 4 or 5 Hz as
seen by a flattening at the noise floor of turbulence spectra
(Carlson 2003). Sampling at 10 Hz captures all signal content
up to 5 Hz. Sampling at a higher frequency than 10 Hz would
not provide any improvement in resolution, and would
merely provide redundant data that were not statistically independent from the 10 Hz data. The instrument has five userselectable velocity range settings. At each measurement point
in a profile, we used the minimum velocity range such that
the mean velocity and fluctuations recorded by the instrument were within the bounds of that range.
Laser Doppler velocimeter—The LDV is a Dantec Dynamics
2-component fiber optic system with a 112 mm diameter
probe, and uses a BSA F60 processor running Flow Software
version 4.0. The probe has an 800 mm focal length with an
expansion ratio of 1.5, producing a measurement volume
0.080 mm in diameter and 1.17 mm in length. The longer
dimension is aligned horizontally, perpendicular to the
mean flow direction. The LDV probe contains both transmitting and receiving optics, and operates in backscatter
mode. It is mounted on a 3D automated traverse that is con-
Accuracy of ADV measurements
trolled through the BSA software. The LDV is driven with 700
mW of argon-ion laser power.
Procedure—LDV velocity measurements rely on light scattered from a measuring volume formed by the intersection of
two or more laser beams. The intersecting beams create a pattern of interference fringes, and the calibration of the instrument is based on the spacing of the fringes. The spacing
depends only on the laser wavelength, which is known, and
the beam crossing angle, which can be accurately measured.
We calibrated the LDV by calculating the beam crossing angle
from measurements of the beam spacing at a given distance
from the intersection at the measurement volume. Error associated with this calibration was less than 1%.
The manufacturer tested the ADV alignment prior to data
acquisition. The instrument output was also calibrated to
match mean free stream velocities reported by our LDV at a
mean flow rate of 14.5 cm/s. SonTek made this adjustment by
redefining the transformation matrix in the configuration file,
and it represented a change of less than 5% relative to the
original factory settings. The configuration file is loaded by
the HorizonADV software and is used for transforming the raw
measurements into an orthogonal coordinate system. The size
and location of the sample volume was determined using the
mapping procedure detailed by Finelli et al. (1999). To do this,
the instrument is systematically lowered over a fine crosshair
while measuring the signal response. The resulting SNR ratio
reported by the instrument varies as a Gaussian with crosshair
position; the peak of the Gaussian corresponds to the sample
volume center. This information was used to calibrate a vertical scale and align the ADV and LDV sampling volumes for
comparative measurements. ADV and LDV measurement locations are always referenced from the center of their sample
volumes (i.e., a measurement location of 2 cm indicates that
the center of the sample volume is located 2 cm from the bed).
The flow tests were conducted at two flow settings for comparison with Spalart’s (1988) data at Reθ = 670 and Reθ = 1410.
Table 1 contains the setup parameters and flow conditions for
each recorded data set. Twenty-minute sample records from
each point in the flow were used to produce profiles of velocity and calculate turbulence statistics. The LDV profiles consisted of 29 points spaced logarithmically from 0.50 mm to
250 mm above the bed. The ADV profiles consisted of 21
Table 1. Flume conditions and flow parameters for the slow and
fast flow cases during operation of each instrument
Slow flow
Static water depth (cm)
Water temperature (°C)
Ufree (cm/s)
U*
ReØ
25
Fast flow
LDV
ADV
LDV
ADV
26
21.75
6.7
0.34
730
26
18.5
6.9
0.34
860
28.3
22
14.6
0.68
1500
28.3
21.5
14.9
0.68
1600
Dombroski and Crimaldi
Accuracy of ADV measurements
Fig. 1. Profiles of (a) signal-to-noise ratio (SNR) and (b) correlation
(COR) for the Reθ = 670 flow case. Profiles represent an average over the
sampling period from the 3 transmit beams of the ADV.
points spaced logarithmically from 3 mm to 240 mm above
the bed. The sampling time (and by extension, the averaging
period) is designed to be sufficient in extent to ensure sufficient statistical convergence. The ADV and LDV were operated
independently because they have different particle seeding
requirements for optimum performance. The seeding particles
used for the ADV are larger than those for the LDV, and their
presence in the flume causes the LDV to report lower peak turbulence statistics. The flow depends only on temperature, flow
depth, and pump settings, the latter two of which we have
precise control over. Temperature effects are very minimal and
are removed in the nondimensionalization scheme by using a
temperature-dependent value for the kinematic viscosity, v.
Although the profiles were not taken simultaneously, the facility provides extremely stable flow conditions over time and
highly repeatable statistics.
There are narrow vertical regions within an ADV profile
where anomalous readings may occur due to acoustic interference with the bed. This occurs when the time for the first
acoustic pulse to travel from the transmitter to the boundary
and back to the transmitter is equal to the pulse lag plus the
time for the second pulse to travel from the transmitter to the
sample volume and back to the transmitter (Carlson 2003).
We avoided sampling in the regions where these interferences
can occur.
Data quality was monitored through SNR and correlation
(COR), which the ADV provides as feedback in real-time as
well as in recorded data. In Fig. 1, we present typical profiles
of SNR and COR, plotted against z (mm), the measurement
height above the bed. The profiles are averaged over the sam-
26
Fig. 2. Sample volume mapping. Vertical axis is the distance from the
ADV transmitter to the crosshairs and horizontal axis is the signal response
from the instrument. Symbols are SNR data from the ADV, and the line is
a least-squares Gaussian fit to the data. The results indicate a sample volume with a height (SVH) of 1.3 cm and centered 4.6 cm from the transmitter (edges defined at e–3 of the peak signal response).
pling period and across the three transmit beams. Although
not presented here, we found the individual response of the
three beams to be quite similar.
Assessment
Sample volume mapping—Sample volume mapping
addresses two issues related to positioning of ADV systems: (a)
quantifying the location of the sample volume in relation to
the transmitter and (b) quantifying the vertical extent of the
sample volume. The procedure provides an independent
method of determining the size and position. The vertical
extent of the sample volume is identified as the region
where the signal strength is large relative to background noise
(Sontek 1997). Fig. 2 displays the signal response as a function
of sample volume position. We chose to define the edges of
the sample volume at e–3 of the peak signal, as this results in a
conservatively large estimate of the vertical extent. For comparative purposes, e–3 corresponds to bounds at 5% of the peak
signal whereas e–2 corresponds to bounds at 14% of the peak
signal. From the mapping procedure, we found the ADV sample volume to be 1.3 cm in vertical extent and centered 4.4 cm
from the probe tip. This compares with nominal measurements of 0.9 cm and 5 cm reported by the manufacturer,
respectively. With a 1.3 cm measuring volume, an instrument
height of 0.7 cm would place the lower edge of the sampling
volume just above the bed surface. Finelli et al. (1999) describe
similar discrepancies between the mapped sample volume size
Dombroski and Crimaldi
Accuracy of ADV measurements
Fig. 3. Dimensional velocity profiles in the streamwise and vertical components for (a) Reθ = 670 and (b) Reθ = 1410 flow cases.
and what the instrument reports. Error in vertical placement
of the sample volume due to user error is estimated to be
within 0.02 cm.
Results
The velocity signal is decomposed according to Ũ =U+u(t),
where U is the mean velocity and u(t) is the fluctuating component. The mean velocity is also described nondimensionally as U+ = U/uτ, where the shear velocity, uτ, is a measure of
shear stress at the bed. It is obtained by fitting the mean velocity profile to the law-of-the-wall, as shown in Fig. 4. The shear
velocity used in nondimensionalized statistics was derived
from the LDV data; however the difference between LDVderived and ADV-derived shear stress values was less than 1%
for the slow flow case and 3% for the fast flow case. Vertical
location is specified in terms of the nondimensional wall unit
z+, defined as z+ = zuτ /v, where z is the distance from the bed
to the center of the sample volume and v is the kinematic viscosity. Fig. 3 contains dimensional velocity data and Figs. 4-6
contain nondimensional velocity, variance, and Reynolds
stress profiles. The figures compare ADV data (closed symbols)
with LDV data (open symbols). Also shown in the figures are
DNS results (solid line) by Spalart (1988). The LDV data are
used as the baseline because it is considered to be the best estimate of true flow characteristics in the flume, because the
technique does not suffer from the same spatial averaging
problems that are inherent to the ADV technique. Streamwise
——
——
and vertical turbulence intensities u2 and w2 , and the Reynolds
——
stress correlation uw, are normalized by u2τ, the square of the
shear velocity. The statistics are plotted versus z+ on the left
axis and z on the right axis. Results are shown for Reθ = 670
(Figs. 3a-6a) and Reθ = 1410 (Figs. 3b-6b) flow cases. For the
Reθ = 670 flow case, the ADV mean velocities were within 5%
of the LDV mean velocities when the center of the sample volume was at or above about 1 cm from the bed (Fig. 3a and Fig. 4a).
For the Reθ = 1410 flow case, 5% accuracy was achieved above
about 0.7 cm from the bed (Fig. 3b and 4b). This vertical location is consistent with our mapping results. A 0.7 cm sampling
height would place the bottom of the measurement volume
(1.3 cm total vertical extent) just above the bed of the flume.
However, for the Reθ = 670 flow case, the ADV results began to
lose accuracy above where the sample volume would be
expected to interfere with the bed. ADV measurements of
mean velocity can be biased due to averaging spatial gradients,
interfering with the static bed, or some combination of the
two. One explanation for the discrepancy is the changing
shape of the velocity boundary layer as Reynolds number
varies. As the shape of the profile changes within the sample
volume, spatial averaging could lead to accuracy differences in
the velocity calculation. Voulgaris and Trowbridge (1998)
investigated this effect by comparing an area-integrated velocity to the instrument’s arithmetic average and found it to be
only on the order of 0.1%, significantly less than the discrepancy reported by our ADV. They attributed velocity underestimation to problems in aligning the sample volumes of the
LDV and ADV. This is a reasonable conclusion given that they
did not conduct a sample volume mapping to verify the size
and location of the sample volume. However, we experienced
27
Dombroski and Crimaldi
Accuracy of ADV measurements
Fig. 4. Nondimensional streamwise velocity profiles for (a) Reθ = 670 and (b) Reθ = 1410 flow cases. The smooth line is direct numerical simulation by
Spalart (1988).
this underestimation of velocity despite having accurately
positioned the instruments. The second potential source of
error in mean velocity measurement is interference with the
bed. At 0.3 cm, the lowest point at which we took data, 25%
of the sample volume is below the bed surface. Correspondingly, mean velocities measured closest to the bed suffer from
significant bias and are underestimated by a factor of two. If
the mapping procedure correctly predicts the vertical extent of
Fig. 5. Nondimensional turbulence intensities for (a) Reθ = 670 and (b) Reθ = 1410 flow cases. The statistic is nondimensionalized by the shear velocity,
u2τ, and is plotted against dimensionless z+ on the left axis (z+ = zuτ/ν) and dimensional z on the right axis (cm). The smooth line is direct numerical simulation (DNS) by Spalart (1988).
28
Dombroski and Crimaldi
Accuracy of ADV measurements
Fig. 6. Nondimensional Reynolds stress for (a) Reθ = 670 and (b) Reθ = 1410 flow cases. The statistic is nondimensionalized by the shear velocity squared,
u2τ, and is plotted against dimensionless z+ on the left axis (z + = zuτ/ν) and dimensional z on the right axis (cm). The smooth line is DNS by Spalart (1988).
the sample volume, then measurements taken above 0.7 cm
should be free from this bias. It is conceivable, however, that
the signal response from a large static object such as the wall
is more pervasive (and therefore extends further) than predicted by a much smaller acoustic target. Finelli et al. (1999)
reported that the mean velocities were accurate to within 1 cm
of the bed for mean freestream flows of approximately 20 cm/s
and 40 cm/s. They then provided some simple calculations
showing that velocity underestimation within 1 cm of the bed
is greater than what would be predicted based on spatial resolution. It is likely that the full extent of errors in mean velocity measurements is due to some combination of factors
related to the sample volume size.
In addition to mean velocities, statistics such as turbulence
intensities and Reynolds stresses are commonly calculated
using ADV velocity records. In Fig. 5, we present streamwise
——
——
(u2 ) and vertical (w2 ) turbulence intensities, normalized by
the shear velocity squared, u2τ, and plotted in terms of wall
units (z+, left-hand axes) and dimensional distance from the
wall (z, right-hand axes). The LDV data are generally in close
agreement with the DNS results for both flow cases. The flow
statistics in the flume (as quantified by the LDV) deviate from
the DNS results in places due to details of the flow in our
flume. In particular, the flow in the flume has a weakly turbulent freestream (with nonzero turbulence intensities),
whereas the DNS results have a laminar freestream (with turbulence intensities asymptoting to zero). So, whereas the DNS
results are useful for verifying that the flow in the flume has
a well-behaved boundary layer, the actual details of the local
flow statistics in the flume are best quantified by the LDV
results. The turbulence intensities calculated from ADV data
follow three broad generalizations: (1) they are artificially
low, (2) the error increases close to the bed, and (3) the magnitude of the errors (and the distance from the wall at which
they appear) is larger for the faster flow case. These generalizations are all consistent with data averaged within a large
sample volume, which (1) spatially averages some energy out
of the flow, (2) is biased by the presence of a solid boundary
close to the sample volume, and (3) averages mean spatial
gradients, which are more severe in faster flows. Near-bed
errors in ADV measurements are well known and have been
reported elsewhere. However, the distance from the wall to
which the errors persist is surprising, especially in the faster
flow case. In this case, the turbulence intensities deviate
markedly from the LDV results as far as 3 to 4 cm from the
wall. In addition, the vertical location of the peak streamwise
intensity value is significantly overestimated by the ADV.
This is due to the ADV measurement volume interfering with
the bed at locations where the LDV and Spalart report the
highest streamwise turbulence levels. The interference causes
values at these locations to be greatly underreported due to
incursion of the static bed into the sample volume. The ADV
is more accurate in estimating the vertical location of the
peak intensity in the w-direction due to the vertical structure
29
Dombroski and Crimaldi
Accuracy of ADV measurements
Fig. 7. Power spectral density (PSD) for the streamwise (Reθ = 1410) flow component of (a) the LDV and ADV at z = 0.7 cm above bed and for (b) the
ADV with noise floor demarcated by a solid black line.
of the boundary layer. The presence of the solid boundary
means that turbulent fluctuations in the vertical direction
peak at larger distances from the bed. ADV measurements can
be made at the higher locations without suffering from interference between the sample volume and the bed. Voulgaris
and Trowbridge (1998) also reported better accuracy in the
vertical direction.
—– ) calculated with ADV
Fig. 6 contains Reynolds stresses (uw
data, again compared with LDV data and DNS results for both
the Reθ = 670 (Fig. 6a) and Reθ= 1410 (Fig. 6b) flow cases. The
LDV data show that the Reynolds stresses in the flume are similar to that predicted by the DNS results, with the exception of
larger values in the freestream, and smaller values near the
Reynolds stress peak. When comparing the Reynolds stresses
calculated from the ADV data with those from the LDV, the
agreement is seen to be very good for distances from the wall
greater than 1 to 2 cm. Below that location, the ADV Reynolds
stresses are underpredicted, and the error is larger closer to the
wall and for the faster flow case.
Spectral analysis—Accurate experimental estimation of turbulence statistics relies on adequate spatial and temporal resolution in the measurement technique, combined with a sufficiently high SNR ratio. Although turbulence statistics are
typically calculated in the time domain for the sake of convenience, an evaluation of the spatial, temporal, and noise
characteristics of data from a particular technique can be more
conveniently examined in the frequency domain. Time series
data are easily converted into the frequency domain using
Fourier techniques, whereby the frequency (or, alternatively,
wavelength) content of a particular statistic can be determined. The power spectral density (PSD) calculation (e.g.,
30
Bendat and Piersol 1984) converts the time history of a statistic into frequency space, showing how contributions to the
variance are distributed with respect to their frequency (or
wavelength) content. Typically, the PSD is normalized such
that it integrates to the variance of the input signal. From the
spectral perspective, spatial and temporal characteristics of the
instrument resolution are more readily observed, as are telltale
signatures of noise content. A typical turbulence spectrum
rolls off at high frequencies due to viscous dissipation at small
spatial scales. White noise, on the other hand, is characterized
by a flat spectrum. A noisy turbulence spectrum, therefore, is
often characterized by a flattening of the spectra at high frequency in the region where the SNR ratio of the instrument
response is very low. Power spectral densities of the streamwise and vertical components of the fast flow case, measured
at a height of 0.7 cm above the bed are displayed in Figs. 7 and 8,
respectively. Plotted in Figs. 7a and 8a are a comparison
between the ADV (gray line) and LDV (black line) data. Figs. 7b
and 8b demonstrate the identification of the ADV noise floor
with a horizontal line. For the present flow conditions and
instrument configurations, the ADV was able to resolve signal
up to about 4-5 Hz. In this case, an ADV sampling rate of 10 Hz
is sufficiently high to capture all signal fluctuations resolved
by the instrument. This noise is inherent to the instrument
and contributes to the variance reported. Thus, theoretically
the true signal is the signal reported from the instrument
minus the noise signal.
We implemented a correction algorithm discussed in Lemmin and Lhermitte (1999) and employed by Voulgaris and
Trowbridge (1998). The method involves subtracting off the
high frequency noise floor in the PSD and was invoked over a
Dombroski and Crimaldi
Accuracy of ADV measurements
Fig. 8. PSD for the vertical (Reθ = 1410) flow component of (a) the LDV and ADV at z = 0.7 cm above bed and for (b) the ADV with noise floor demarcated by a solid black line.
subset of data from both instruments. The effect of this filtering depends on the magnitude of the noise floor relative to the
signal at that location. The LDV signal has a very small noise
component and therefore filtering does not have a significant
effect on the LDV data. The effect of filtering ADV data are
more considerable due to the higher noise levels inherent to
acoustic instruments. A greater portion of the instrument
response is noise and the effect of removing it is to reduce the
energy content in the data reported from the instrument. The
unfiltered ADV data already underreports turbulence levels in
the flow prior to filtering, which means that removal of noise
brings the data farther from the LDV and Spalart data. Presented in Fig. 9 are filtered (solid circles) and unfiltered (open
circles) ADV variances plotted with LDV data (open squares)
for comparison. The variance profiles in Fig. 9 correspond to
the streamwise component of the Reθ = 1410 flow case and are
calculated in the spectral domain. The figure shows that the
error in the ADV results is even greater if the contributions to
the variance from the instrument noise are removed.
Spectral comparison between LDV and ADV data indicates
that the large sample volume is not solely responsible for
underreported turbulence statistics. The energy cascade model
predicts a range of eddy sizes from large, low frequency turbulent motions progressing down to small, high frequency
motions at the Kolmogorov microscale. Thus, in a spectral representation of flow characteristics, there is a relationship
between frequency and eddy size. The ADV technique cannot
be expected to resolve scales of motions that are on the order
of the sample volume size and smaller. If the ADV were only
underreporting energy at scales smaller than the sample volume, then the PSD would drop off at a frequency correspon-
ding to a length-scale on the order of the sample volume size.
However, the magnitude of the ADV PSD is smaller than that
of the LDV over the full frequency range of the plot (excepting only the lowest frequencies). Therefore, energy is being
Fig. 9. Filtered and unfiltered ADV turbulence intensities compared with
LDV and Spalart data for the Reθ = 1410 flow case. The statistic is nondimensionalized by the shear velocity squared, u2τ, and is plotted against
dimensionless z+ on the left axis (z+ = zuτ/ν) and dimensional z on the right
axis (cm). The line is DNS by Spalart (1988).
31
Dombroski and Crimaldi
Accuracy of ADV measurements
underreported over a wide range of spatial scales, even those
much larger than the sample volume.
Discussion
ADV performance is evaluated by comparing results with an
independent measure of known flow statistics. We found that,
in general, ADV employed in turbulent boundary layers is not
as accurate as previously reported, and the region where inaccurate results are obtained extends farther from the bed than
previously reported. We also give results from ADV sample volume mapping in which we found the measurement volume to
be larger and located differently than specified by the manufacturer. This finding is consistent with previous studies. The
results from the mapping procedure provide the means for
accurate placement of the measuring region. However, locating
the lower edge of the sample volume above the bed is not a sufficient condition to guarantee accurate results. At 0.7 cm above
the bed, the ADV measured mean velocities to within 5% accuracy only for the Re = 1410 flow case. For the Re = 670 flow
case, 5% accuracy was achieved only to within 1 cm of the bed.
Results for turbulence statistics show that ADV inaccuracies
can extend much farther from the bed, up to about 4 cm in
some cases. Because the nature of these inaccuracies is to
underreport statistics, the high noise level inherent to ADV
contributes to the signal in a way that makes reported turbulence statistics appear better than they would without the addition of noise to the signal. Therefore, we conclude that noise
filtering is not a beneficial or worthwhile approach to improving accuracy of results from the ADV technique.
Comments and recommendations
The ADV technique is limited largely due to a sample volume that is (1) larger than the smallest scales of motion in
most turbulent flows and (2) large compared to the distances
from the bed where near-wall measurements are commonly
required. Furthermore, ADV measurements can be biased at
distances from the bed that are outside the theoretical and
experimental extent of the sample volume. Interference
between the bed and sample volume causes the SNR reported
by the instrument to increase in proportion to the amount of
interference, which can give users a false level of confidence in
results. This point underscores the importance of conducting
sample volume mapping before data acquisition. Mean velocities can be accurately measured to within 1 cm of the bed. ADV
users should exercise caution, however, when reporting turbulence statistics. These results indicate that the accuracy within
the 1 to 3 cm range is somewhat dependent on the flow itself.
For slower flows, turbulence statistics may be accurate to as
close as 1 cm from the bed, however for faster flows the same
level of accuracy may only be possible 3 to 4 cm from the bed.
Further research is needed to provide an objective evaluation of
ADV performance while taking measurements in turbulent
flows with a rough substrate, as this will add additional complexity in aligning the ADV and interpreting results.
32
References
Albrecht, H.- E., M. Borys, N. Damaschke, and C. Tropea. 2003.
Laser Doppler and phase Doppler measurement techniques.
Springer.
Adrian, R. J., and B. J. Thompson. 1993. Selected papers on
laser Doppler velocimetry. SPIE Optical Engineering Press.
Bendat, J. S., and A. G. Piersol. 1984. Random data: analysis
and measurement procedures, 2nd ed. John Wiley & Sons.
Butt, T., P. Russell, J. Puleo, J. Miles, and G. Masselink. 2004.
The influence of bore turbulence on sediment transport in
the swash and inner surf zones. Cont. Shelf. Res. 24(7-8):
757-771.
Carbonneau, P. E., and N. E. Bergeron. 2000. The effect of bedload transport on mean and turbulent flow properties. Geomorph. 35:267-278.
Carlson, E. M. 2003. An evaluation of acoustic Doppervelocimetry in laboratory flumes and natural streams. Master thesis. Univ. of Colorado at Boulder.
Carollo, F. G., V. Ferro, and D. Termini. 2002. Flow velocity
measurements in vegetated channels. J. Hydr. Eng. 128:
664-673.
Cheng, N. S., and Y. M. Chiew. 1998. Turbulent open-channel
flow with upward seepage. J. Hydr. Res. 36:415-431.
Cornelisen, C. D., and F. I. M. Thomas. 2004. Ammonium and
nitrate uptake by leaves of the seagrass Thalassia testudinum: impact of hydrodynamic regime and epiphyte
cover on uptake rates. J. Mar. Sys. 49(1-4):177-194.
Crimaldi, J. P., M. B. Wiley, and J. R. Koseff. 2002. The relationship between mean and instantaneous structure in turbulent passive scalar plumes. J. Turb. 3(014):2-24.
Doering, J. C., and A. J. Baryla. 2002. An investigation of the
velocity field under regular and irregular waves over a sand
beach. Coast. Eng. 44:275-300.
Eckman, J. E., W. B. Savidge, and T. F. Gross. 1990. Relationship between duration of cyprid attachment and drag
forces associated with detachment of Balanus amphitrite
cyprids. Mar. Biol. 107:111-118.
Finelli, C. M., D. D. Hart, D. M. Fonseca. 1999. Evaluating the
spatial resolution of an acoustic Doppler velocimeter and
the consequences for measuring near-bed flows. Limnol.
Oceanogr. 44:1793-1801.
Friedrichs, M., G. Graf, and B. Springer. 2000. Skimming flow
induced over a simulated polychaete tube lawn at low population densities. Marine Ecology Progress Series. 192:219-228.
Kim, S. C., C. T. Friedrichs, J. P. Y. Maa, and L. D. Wright. 2000.
Estimating bottom stress in tidal boundary layer from
acoustic Doppler velocimeter data. J. Hydr. Eng. 126:399-406.
Kraus, N. C., A. Lohrmann, and R. Cabrera. 1994. New
acoustic meter for measuring 3D laboratory flows. J. Hydr.
Eng. 120(3):406-412.
Lane, S. N., K. F. Bradbrook, S. W. B. Caudwell, and K. S. Richards.
1998. Mixing processes at river confluences: field informed
numerical modeling, p. 345-350. In J. H. W. Lee, A. W. Jayawardena, and Z. Y. Wang [eds.], Environmental hydraulics. Balkema.
Dombroski and Crimaldi
Lawless, M., and A. Robert. 2001. Scales of boundary resistance
in coarse-grained channels: turbulent velocity profiles and
implications. Geomorph. 39:221-238.
Lemmin, U., and R. Lhermitte. 1999. Discussion: ADV measurements of turbulence: can we improve their interpretation? J. Hydr. Eng. 125(9):987-988.
Lohrmann, A., R. Cabrera, and N. C. Kraus. 1994. AcousticDoppler velocimeter (ADV) for laboratory use, p. 351-365.
In Proc. conf. on fundamentals and advancements in
hydraulic measurements and experimentation. American
Society of Civil Engineers.
Mutz, M. 2000. Influences of woody debris on flow patterns
and channel morphology in a low energy, sand-bed stream
reach. Int. Rev. Hydrobiol. 85:107-121.
Nezu, I., and W. Rodi. 1986. Open channel flow measurements with a laser Doppler anemometer. J. Hydraul.
Eng.. 112: 335-355.
Nikora, V. I., and D. G. Goring. 2000. Flow turbulence over fixed
and weakly mobile gravel beds. J. Hydr. Eng. 126(9):679-690.
Osborne, P. D., K. Black, P. V. Villard, and S. Douglass. 1997.
Field measurements of 3-D velocity structure and suspended sediments under locally generated waves and swell.
Pacific Coasts and Ports 1997. 564.
Accuracy of ADV measurements
SonTek. 1997. Sontek ADVField acoustic Doppler velocimeter: technical documentation. version 4.0. San Diego, CA:
SanTek.
Spalart, P. R. 1988. Direct simulation of a turbulent boundary
layer up to Rθ = 1410. J. Fluid Mech. 187:61-98.
Talke, S. A., and M. T. Stacey. 2003. The influence of oceanic
swell on flows over an estuarine intertidal mudflat in San
Francisco Bay. Est. Coast. Shelf Sci. 58(3):541-554.
Tritico, H. M., and R. H. Hotchkiss. 2005. Unobstructed and
obstructed turbulent flow in gravel bed rivers. J. Hydr Eng.
131(8):635-645.
Venditti, J. G., and B. O. Bauer. 2005. Turbulent flow over a
dune: Green River, Colorado. E. Surf. Proc. Land. 30(3):
289-304.
Voulgaris, G., and J. H. Trowbridge. 1998. Evaluation of the
acoustic Doppler velocimeter (ADV) for turbulence measurements. J. Atmos. Ocean. Technol. 15:272-289.
Yahel, R., G. Yahel, and A. Genin. 2002. Daily cycles of suspended sand at coral reefs: a biological control. Limnol.
Oceanogr. 47(4):1071-1083.
Submitted 16 March 2006
Revised 20 September 2006
Accepted 12 October 2006
33